Experiment Tracking#
Open MODEL TRAINING → Tasks, Models, Datasets, or Artifacts in OptScale AI to run and compare training workloads, register models, version datasets, and store run outputs. These pages provide experiment lineage and lifecycle management before infrastructure pages such as Shared Environments. For the full Model Training menu, see Model Training overview; for AI Gateway and FinOps pages above this section, see Core Services.
Tasks#
Open MODEL TRAINING → Tasks to create and monitor ML and AI workloads. A task groups related runs so you can compare executions, review metrics, and track experiment history in one place.
The Tasks page provides controls for creating, filtering, and monitoring ML training and pipeline executions.
Use REFRESH to reload task and run data. PROFILING INTEGRATION opens guidance for connecting training code with the profiling SDK, while EXECUTORS displays instances and environments used by recent runs.
Use + ADD to register a new task before the first profiled execution. Use Search, Owner, Status, and Goal filters to narrow the task list.
The main table displays task information, including the task name, owner, latest execution status, timing, and collected profiling metrics. Select a task name to open the detail page with runs, metrics, and profiling data.
Use the task detail page to compare runs, inspect metrics, and manage access. Related outputs appear under Artifacts; trained models under Models.
Models#
Open MODEL TRAINING → Models for the OptScale AI model registry—centralized storage for trained models, versions, aliases, and metadata used in production and experimentation.
The Models page provides controls for managing the model registry and registered model versions.
Use REFRESH to reload the model catalog, + ADD to register a new model, and Search to find models by name, tag, or other visible fields.
The main table displays model information, including the model name, registered version count, deployment aliases, tags, and available management actions. Select a model name to open the detail page for versions, aliases, and metadata.
On the model detail page, review version history, aliases, tags, and links to related Tasks and Datasets.
Datasets#
Open MODEL TRAINING → Datasets to manage dataset catalog entries, versions, and lineage across training and evaluation workflows in OptScale AI.
The Datasets page provides controls for browsing, filtering, and registering datasets and dataset versions.
Use Count to view the number of datasets in the current result set. Use REFRESH to reload catalog data. LABEL, USED IN, and PRODUCED BY filters help narrow datasets by lineage and workflow usage.
Use + ADD to register a new dataset or dataset version, and use Search to filter datasets by name, key, or description.
The main table displays dataset identity, registration time, description, version history, lineage relationships, and organizational labels.
Footer counters show total and Displayed rows for the active filters.
Artifacts#
Open MODEL TRAINING → Artifacts to browse run outputs from training and pipeline executions—checkpoints, plots, logs, environment files, and reports. Artifacts are tied to a run and often reference external storage (for example Amazon S3).
The Artifacts page provides controls for browsing, filtering, and managing run outputs and generated files.
Use the time-range filters (All, 1 day, 1 week, 2 weeks, 1 month, or Custom) to narrow artifacts by creation date. Use TASK (ANY) to filter artifacts by related task, Search to filter by name, path, tags, or description, and REFRESH to reload the artifact list.
The main table displays artifact information, including the artifact name, storage path, description, related run, creation time, tags, and available management actions.
Follow the Run link to inspect metrics and related Tasks entries.
Related documentation#
| Topic | Where to read more |
|---|---|
| Environments, cloud, schedules, CI | Environments & Operations |
| Model Training menu overview | Model Training |
| AI Gateway, policies, FinOps | Core Services |
| Platform architecture | Architecture Overview |